@react-x11/components 0.7.1 → 0.9.0

package/src/rich-text-editor/markdown.ts

@@ -0,0 +1,653 @@
+// Markdown in and out of the editor's document, through the markdown AST.
+//
+// `src/internal/markdown/parse.ts` is `<Markdown>`'s own parser — GFM, with
+// the same tolerance and the same deliberate deviations — so a note written
+// in the editor and one rendered by `<Markdown>` agree about what the text
+// means, and this file is only the translation between two trees: the AST's
+// nested inline nodes and ProseMirror's flat mark sets. Going out, the flat
+// marks are regrouped into the nesting that needs the fewest delimiters, and
+// `stringify.ts` writes it.
+//
+// It speaks the reference schema's names (`./schema.ts`) — and TipTap's
+// camelCase ones as well, so a schema built from TipTap's extensions round-
+// trips through markdown without a codec of its own. A node neither knows is
+// kept as what it contains: a textblock's text, a container's blocks. A
+// schema with nodes markdown cannot say should pass a codec (`markdown`) or
+// use `format="html"`, which goes through the schema's own `toDOM`.
+import type {
+  Attrs,
+  Mark,
+  MarkType,
+  Node as PMNode,
+  NodeType,
+  Schema,
+} from 'prosemirror-model';
+
+import { parse } from '../internal/markdown/parse.js';
+import { stringify } from '../internal/markdown/stringify.js';
+import type {
+  BlockNode,
+  InlineNode,
+  LinkInline,
+  ListItem,
+  TableAlign,
+} from '../internal/markdown/ast.js';
+
+/** How a document becomes markdown and back — the default one, or an app's
+ *  for a schema this one does not understand. */
+export interface MarkdownCodec {
+  parse(source: string): PMNode;
+  serialize(doc: PMNode): string;
+}
+
+/** The codec for `schema`: the reference names, TipTap's, or both. */
+export function markdownCodec(schema: Schema): MarkdownCodec {
+  return {
+    parse: (source) => docFromMarkdown(schema, source),
+    serialize: (doc) => markdownFromDoc(doc),
+  };
+}
+
+// --- names -------------------------------------------------------------------
+
+/** The spellings each thing goes by: ProseMirror's reference schemas first,
+ *  then TipTap's. */
+const NODE_NAMES = {
+  paragraph: ['paragraph'],
+  heading: ['heading'],
+  code: ['code_block', 'codeBlock'],
+  quote: ['blockquote'],
+  bullet: ['bullet_list', 'bulletList'],
+  ordered: ['ordered_list', 'orderedList'],
+  task: ['task_list', 'taskList'],
+  item: ['list_item', 'listItem'],
+  taskItem: ['task_item', 'taskItem'],
+  rule: ['horizontal_rule', 'horizontalRule'],
+  table: ['table'],
+  row: ['table_row', 'tableRow'],
+  cell: ['table_cell', 'tableCell'],
+  header: ['table_header', 'tableHeader'],
+  image: ['image'],
+  hardBreak: ['hard_break', 'hardBreak'],
+} as const;
+
+const MARK_NAMES = {
+  strong: ['strong', 'bold'],
+  em: ['em', 'italic'],
+  del: ['strike', 's', 'del', 'strikethrough'],
+  code: ['code'],
+  link: ['link'],
+} as const;
+
+type NodeRole = keyof typeof NODE_NAMES;
+type MarkRole = keyof typeof MARK_NAMES;
+
+const NODE_ROLE = new Map<string, NodeRole>();
+for (const [role, names] of Object.entries(NODE_NAMES)) {
+  for (const name of names) NODE_ROLE.set(name, role as NodeRole);
+}
+const MARK_ROLE = new Map<string, MarkRole>();
+for (const [role, names] of Object.entries(MARK_NAMES)) {
+  for (const name of names) MARK_ROLE.set(name, role as MarkRole);
+}
+
+function nodeType(schema: Schema, role: NodeRole): NodeType | undefined {
+  for (const name of NODE_NAMES[role]) {
+    const type = schema.nodes[name];
+    if (type) return type;
+  }
+  return undefined;
+}
+
+function markType(schema: Schema, role: MarkRole): MarkType | undefined {
+  for (const name of MARK_NAMES[role]) {
+    const type = schema.marks[name];
+    if (type) return type;
+  }
+  return undefined;
+}
+
+/** Only the attributes `type` declares — TipTap's `orderedList` calls the
+ *  start `start` where prosemirror-markdown calls it `order`, and a node
+ *  should not be handed a name it does not have. */
+function attrsFor(type: NodeType, wanted: Record<string, unknown>): Attrs {
+  const out: Record<string, unknown> = {};
+  const spec = type.spec.attrs ?? {};
+  for (const [name, value] of Object.entries(wanted)) {
+    if (name in spec) out[name] = value;
+  }
+  return out;
+}
+
+// --- markdown → document -----------------------------------------------------
+
+/** Parse markdown into a document of `schema`. Never throws on content: what
+ *  the schema cannot hold is kept as the nearest thing it can. */
+export function docFromMarkdown(schema: Schema, source: string): PMNode {
+  const ast = parse(source, { partial: false });
+  const blocks = blocksToNodes(schema, ast.blocks);
+  return (
+    schema.topNodeType.createAndFill(null, blocks) ??
+    schema.topNodeType.createAndFill()!
+  );
+}
+
+function blocksToNodes(schema: Schema, blocks: readonly BlockNode[]): PMNode[] {
+  const out: PMNode[] = [];
+  for (const block of blocks) out.push(...blockToNodes(schema, block));
+  return out;
+}
+
+function paragraph(schema: Schema, content: PMNode[]): PMNode {
+  const type =
+    nodeType(schema, 'paragraph') ??
+    schema.topNodeType.contentMatch.defaultType!;
+  return type.createAndFill(null, content) ?? type.create();
+}
+
+function blockToNodes(schema: Schema, block: BlockNode): PMNode[] {
+  switch (block.type) {
+    case 'paragraph':
+      return [paragraph(schema, inlineToNodes(schema, block.children, []))];
+
+    case 'heading': {
+      const type = nodeType(schema, 'heading');
+      const content = inlineToNodes(schema, block.children, []);
+      if (!type) return [paragraph(schema, content)];
+      return [
+        type.createAndFill(attrsFor(type, { level: block.depth }), content) ??
+          paragraph(schema, content),
+      ];
+    }
+
+    case 'code': {
+      const type = nodeType(schema, 'code');
+      const text = block.text ? [schema.text(block.text)] : [];
+      if (!type) return [paragraph(schema, text)];
+      const info = block.info ?? block.lang;
+      return [
+        type.create(
+          attrsFor(type, {
+            params: info,
+            language: block.lang || null,
+          }),
+          text,
+        ),
+      ];
+    }
+
+    case 'quote': {
+      const type = nodeType(schema, 'quote');
+      const children = blocksToNodes(schema, block.children);
+      if (!type) return children;
+      return [
+        type.createAndFill(
+          null,
+          children.length ? children : [paragraph(schema, [])],
+        ) ?? paragraph(schema, []),
+      ];
+    }
+
+    case 'list':
+      return listToNodes(
+        schema,
+        block.ordered,
+        block.start,
+        block.tight,
+        block.items,
+      );
+
+    case 'table': {
+      const table = nodeType(schema, 'table');
+      const row = nodeType(schema, 'row');
+      const cell = nodeType(schema, 'cell');
+      const header = nodeType(schema, 'header') ?? cell;
+      if (!table || !row || !cell || !header) {
+        // no tables here: a row per paragraph, cells kept apart by a bar
+        return [block.header, ...block.rows].map((cells) =>
+          paragraph(
+            schema,
+            cells.flatMap((c, i) => [
+              ...(i > 0 ? [schema.text(' | ')] : []),
+              ...inlineToNodes(schema, c, []),
+            ]),
+          ),
+        );
+      }
+      const makeRow = (cells: InlineNode[][], type: NodeType): PMNode =>
+        row.create(
+          null,
+          cells.map(
+            (c, i) =>
+              type.createAndFill(
+                attrsFor(type, { align: block.align[i] ?? null }),
+                inlineToNodes(schema, c, []),
+              ) ?? type.createAndFill()!,
+          ),
+        );
+      return [
+        table.create(null, [
+          makeRow(block.header, header),
+          ...block.rows.map((r) => makeRow(r, cell)),
+        ]),
+      ];
+    }
+
+    case 'rule': {
+      const type = nodeType(schema, 'rule');
+      return type ? [type.create()] : [];
+    }
+
+    case 'component':
+      return blocksToNodes(schema, block.children);
+  }
+}
+
+function listToNodes(
+  schema: Schema,
+  ordered: boolean,
+  start: number,
+  tight: boolean,
+  items: readonly ListItem[],
+): PMNode[] {
+  const isTask =
+    !ordered && items.length > 0 && items.every((i) => i.checked !== null);
+  // TipTap keeps task items in a list type of their own; the reference
+  // schema (and this one) keeps them in a bullet list, as GFM does
+  const taskList = isTask ? nodeType(schema, 'task') : undefined;
+  const listType = taskList ?? nodeType(schema, ordered ? 'ordered' : 'bullet');
+  const itemType =
+    (taskList ? nodeType(schema, 'taskItem') : undefined) ??
+    nodeType(schema, 'item');
+  if (!listType || !itemType) {
+    return items.flatMap((item) => blocksToNodes(schema, item.children));
+  }
+  const nodes = items.map((item) => {
+    let children = blocksToNodes(schema, item.children);
+    // an item's first child must be a paragraph (`paragraph block*`); one
+    // that opened with a fence or a nested list gets an empty paragraph in
+    // front of it, which is what that markdown looks like anyway
+    const first = itemType.contentMatch.defaultType;
+    if (first && (children.length === 0 || children[0].type !== first)) {
+      children = [first.createAndFill()!, ...children];
+    }
+    return (
+      itemType.createAndFill(
+        attrsFor(itemType, { checked: item.checked }),
+        children,
+      ) ?? itemType.createAndFill()!
+    );
+  });
+  return [
+    listType.create(attrsFor(listType, { order: start, start, tight }), nodes),
+  ];
+}
+
+function withMark(
+  marks: readonly Mark[],
+  type: MarkType | undefined,
+  attrs?: Attrs,
+): readonly Mark[] {
+  return type ? type.create(attrs).addToSet(marks) : marks;
+}
+
+function plainOf(nodes: readonly InlineNode[]): string {
+  let out = '';
+  for (const node of nodes) {
+    if (node.type === 'text' || node.type === 'code') out += node.text;
+    else if (node.type === 'break') out += ' ';
+    else if ('children' in node) out += plainOf(node.children);
+  }
+  return out;
+}
+
+function inlineToNodes(
+  schema: Schema,
+  nodes: readonly InlineNode[],
+  marks: readonly Mark[],
+): PMNode[] {
+  const out: PMNode[] = [];
+  for (const node of nodes) {
+    switch (node.type) {
+      case 'text':
+        if (node.text) out.push(schema.text(node.text, marks));
+        break;
+      case 'code':
+        if (node.text) {
+          out.push(
+            schema.text(node.text, withMark(marks, markType(schema, 'code'))),
+          );
+        }
+        break;
+      case 'strong':
+        out.push(
+          ...inlineToNodes(
+            schema,
+            node.children,
+            withMark(marks, markType(schema, 'strong')),
+          ),
+        );
+        break;
+      case 'em':
+        out.push(
+          ...inlineToNodes(
+            schema,
+            node.children,
+            withMark(marks, markType(schema, 'em')),
+          ),
+        );
+        break;
+      case 'del':
+        out.push(
+          ...inlineToNodes(
+            schema,
+            node.children,
+            withMark(marks, markType(schema, 'del')),
+          ),
+        );
+        break;
+      case 'link':
+        out.push(...linkToNodes(schema, node, marks));
+        break;
+      case 'break': {
+        const type = nodeType(schema, 'hardBreak');
+        if (type) out.push(type.create(null, null, marks));
+        else out.push(schema.text(' ', marks));
+        break;
+      }
+      case 'expression':
+        out.push(schema.text(`{${node.src}}`, marks));
+        break;
+      case 'component':
+        out.push(...inlineToNodes(schema, node.children, marks));
+        break;
+    }
+  }
+  return out;
+}
+
+function linkToNodes(
+  schema: Schema,
+  node: LinkInline,
+  marks: readonly Mark[],
+): PMNode[] {
+  if (node.image) {
+    const type = nodeType(schema, 'image');
+    const alt = plainOf(node.children);
+    if (type && node.href !== null) {
+      return [
+        type.create(
+          attrsFor(type, {
+            src: node.href,
+            alt: alt || null,
+            title: node.title ?? null,
+          }),
+          null,
+          marks,
+        ),
+      ];
+    }
+    return alt ? [schema.text(alt, marks)] : [];
+  }
+  const link = markType(schema, 'link');
+  const inner =
+    node.href !== null && link
+      ? withMark(marks, link, { href: node.href, title: node.title ?? null })
+      : marks;
+  return inlineToNodes(schema, node.children, inner);
+}
+
+// --- document → markdown -----------------------------------------------------
+
+/** Write a document of the reference (or TipTap's) schema as markdown. */
+export function markdownFromDoc(doc: PMNode): string {
+  return stringify(blocksOf(doc));
+}
+
+function blocksOf(parent: PMNode): BlockNode[] {
+  const out: BlockNode[] = [];
+  parent.forEach((child) => out.push(...nodeToBlocks(child)));
+  return out;
+}
+
+function nodeToBlocks(node: PMNode): BlockNode[] {
+  switch (NODE_ROLE.get(node.type.name)) {
+    case 'paragraph':
+      return [{ type: 'paragraph', children: inlineOf(node) }];
+    case 'heading':
+      return [
+        {
+          type: 'heading',
+          depth: Number(node.attrs.level ?? 1),
+          children: inlineOf(node),
+        },
+      ];
+    case 'code': {
+      const info = String(
+        node.attrs.params ?? node.attrs.language ?? '',
+      ).trim();
+      const lang = (info.split(/[ \t]/, 1)[0] ?? '').toLowerCase();
+      return [
+        {
+          type: 'code',
+          lang,
+          ...(info !== lang ? { info } : null),
+          text: node.textContent,
+          closed: true,
+        },
+      ];
+    }
+    case 'quote':
+      return [{ type: 'quote', children: blocksOf(node) }];
+    case 'bullet':
+    case 'ordered':
+    case 'task': {
+      const ordered = NODE_ROLE.get(node.type.name) === 'ordered';
+      const items: ListItem[] = [];
+      node.forEach((item) => {
+        const checked = item.attrs.checked;
+        items.push({
+          checked:
+            typeof checked === 'boolean'
+              ? checked
+              : NODE_ROLE.get(item.type.name) === 'taskItem'
+                ? false
+                : null,
+          children: blocksOf(item),
+        });
+      });
+      return [
+        {
+          type: 'list',
+          ordered,
+          start: Number(node.attrs.order ?? node.attrs.start ?? 1),
+          tight: node.attrs.tight !== false,
+          items,
+        },
+      ];
+    }
+    case 'rule':
+      return [{ type: 'rule' }];
+    case 'table':
+      return [tableOf(node)];
+    default:
+      if (node.isTextblock)
+        return [{ type: 'paragraph', children: inlineOf(node) }];
+      if (!node.isLeaf) return blocksOf(node);
+      return [];
+  }
+}
+
+function tableOf(table: PMNode): BlockNode {
+  const grid: InlineNode[][][] = [];
+  const align: TableAlign[] = [];
+  table.forEach((row) => {
+    const cells: InlineNode[][] = [];
+    row.forEach((cell) => {
+      const col = cells.length;
+      if (align[col] === undefined) {
+        const a = cell.attrs.align;
+        align[col] = a === 'left' || a === 'center' || a === 'right' ? a : null;
+      }
+      cells.push(cellInline(cell));
+    });
+    grid.push(cells);
+  });
+  const width = Math.max(1, ...grid.map((r) => r.length));
+  const pad = (r: InlineNode[][]): InlineNode[][] =>
+    Array.from({ length: width }, (_, i) => r[i] ?? []);
+  for (let i = 0; i < width; i++) align[i] ??= null;
+  return {
+    type: 'table',
+    align: align.slice(0, width),
+    header: pad(grid[0] ?? []),
+    rows: grid.slice(1).map(pad),
+  };
+}
+
+/** A cell's content as one line: its own inline content, or — for a schema
+ *  whose cells hold blocks — each textblock's, a space between. */
+function cellInline(cell: PMNode): InlineNode[] {
+  if (cell.isTextblock) return inlineOf(cell);
+  const out: InlineNode[] = [];
+  cell.descendants((node) => {
+    if (!node.isTextblock) return true;
+    if (out.length) out.push({ type: 'text', text: ' ' });
+    out.push(...inlineOf(node));
+    return false;
+  });
+  return out;
+}
+
+interface OpenMark {
+  mark: Mark;
+  children: InlineNode[];
+}
+
+/**
+ * A textblock's flat marked runs, as the AST's nested inline tree.
+ *
+ * The open marks are a stack. At each child, every mark the child no longer
+ * carries is closed — with everything opened inside it, to be reopened —
+ * and the marks it adds are opened, **the longest-running outermost**, so
+ * that "bold, then bold italic, then bold" is one bold around an italic and
+ * not two bolds either side of one: two runs of `**` meeting reads back as
+ * one run of four. Code is never a container: it is the leaf, innermost,
+ * where markdown needs it.
+ */
+function inlineOf(block: PMNode): InlineNode[] {
+  const children: PMNode[] = [];
+  block.forEach((child) => children.push(child));
+  const containers = children.map((child) =>
+    child.marks.filter((m) => {
+      const role = MARK_ROLE.get(m.type.name);
+      return role !== undefined && role !== 'code';
+    }),
+  );
+  /** The last index, from `i` on, through which `mark` runs unbroken. */
+  const runsTo = (i: number, mark: Mark): number => {
+    let j = i;
+    while (j + 1 < children.length && mark.isInSet(containers[j + 1])) j++;
+    return j;
+  };
+
+  const root: InlineNode[] = [];
+  const stack: OpenMark[] = [];
+  const top = (): InlineNode[] =>
+    stack.length ? stack[stack.length - 1].children : root;
+
+  children.forEach((child, i) => {
+    const want = containers[i];
+    let keep = 0;
+    while (keep < stack.length && stack[keep].mark.isInSet(want)) keep++;
+    stack.length = keep;
+    // Longest-running first; between two that run as far, emphasis outside
+    // a link — `~~[ a](u)~~` can hold the space at the link's edge, while
+    // `[~~ a~~](u)` would have to move it out of the strikethrough.
+    const opening = want
+      .filter((m) => !stack.some((open) => open.mark.eq(m)))
+      .map((m, order) => ({
+        m,
+        end: runsTo(i, m),
+        link: MARK_ROLE.get(m.type.name) === 'link' ? 1 : 0,
+        order,
+      }))
+      .sort((a, b) => b.end - a.end || a.link - b.link || a.order - b.order);
+    for (const { m } of opening) {
+      const inner: InlineNode[] = [];
+      top().push(containerOf(m, inner));
+      stack.push({ mark: m, children: inner });
+    }
+    const code = child.marks.some((m) => MARK_ROLE.get(m.type.name) === 'code');
+    top().push(...leafOf(child, code));
+  });
+  return root;
+}
+
+function containerOf(mark: Mark, children: InlineNode[]): InlineNode {
+  switch (MARK_ROLE.get(mark.type.name)) {
+    case 'strong':
+      return { type: 'strong', children };
+    case 'em':
+      return { type: 'em', children };
+    case 'del':
+      return { type: 'del', children };
+    default: {
+      const title = mark.attrs.title;
+      return {
+        type: 'link',
+        href: String(mark.attrs.href ?? ''),
+        image: false,
+        children,
+        ...(typeof title === 'string' && title ? { title } : null),
+      };
+    }
+  }
+}
+
+function leafOf(node: PMNode, code: boolean): InlineNode[] {
+  if (node.isText) {
+    const text = node.text ?? '';
+    return [code ? { type: 'code', text } : { type: 'text', text }];
+  }
+  const role = NODE_ROLE.get(node.type.name);
+  if (role === 'hardBreak') return [{ type: 'break' }];
+  if (role === 'image') {
+    const alt = node.attrs.alt;
+    const title = node.attrs.title;
+    return [
+      {
+        type: 'link',
+        href: String(node.attrs.src ?? ''),
+        image: true,
+        children:
+          typeof alt === 'string' && alt ? [{ type: 'text', text: alt }] : [],
+        ...(typeof title === 'string' && title ? { title } : null),
+      },
+    ];
+  }
+  // an inline node markdown has no word for: its text, if it has any
+  const leaf = node.type.spec.leafText?.(node) ?? node.textContent;
+  return leaf ? [{ type: 'text', text: leaf }] : [];
+}
+
+// --- plain text ----------------------------------------------------------------
+
+/** One textblock per line, both ways — `format="text"`. */
+export function docFromText(schema: Schema, text: string): PMNode {
+  const lines = text.replace(/\r\n?/g, '\n').split('\n');
+  return (
+    schema.topNodeType.createAndFill(
+      null,
+      lines.map((line) => paragraph(schema, line ? [schema.text(line)] : [])),
+    ) ?? schema.topNodeType.createAndFill()!
+  );
+}
+
+export function textFromDoc(doc: PMNode): string {
+  return doc.textBetween(0, doc.content.size, '\n', (leaf) =>
+    NODE_ROLE.get(leaf.type.name) === 'hardBreak'
+      ? '\n'
+      : (leaf.type.spec.leafText?.(leaf) ?? ''),
+  );
+}